Alice · extension
The web researcher that treats safety as a feature.
Alice collects text and images from the web and hands you a clean dataset you can prove the provenance of. She is deliberately narrow: she visits only a short, curated list of reputable, well-licensed sources — Wikipedia, Wikimedia Commons, Project Gutenberg, Poly Haven, ambientCG — and never crawls the open web. Alice runs as an extension on the godspeed engine, and every conjure subscription includes her free.
Buy godspeed + alice
Two license keys arrive by email — one for the engine, one for alice. Buying conjure instead? Alice rides along at no charge.
What it does
Never touches the open web
Eleven allowlisted hosts, checked before any request is made. That restriction is the safety control, chosen because the alternative has a documented history: the indiscriminate open-web scrape behind LAION-5B was found to contain material that forced the dataset to be withdrawn.
Refuses a licence it does not recognise
Per file, not per site: CC0, public domain, CC BY and CC BY-SA are kept;
-NC, -ND and anything unrecognised are skipped. Unknown is a
refusal, not a default-yes.
Every row is traceable
Source URL, host, fetch time, SHA-256 and licence ride every item, and export is a
flat dataset.jsonl your trainer reads directly — so you can filter
on rights before you train, and prove where a row came from afterwards.
Feeds the AI line
Alice is the acquisition layer our own models train on — the same pipeline that feeds conjure’s live refinement loop is the one you get.
Deeper dive: the alice documentation.